The VaR historical method is the most intuitive approach to estimating Value at Risk. Rather than assuming a statistical distribution for portfolio returns, the historical method uses actual past returns to estimate how much you could lose on a bad day. This guide covers how the historical method works, its formula, a step-by-step example, and how to implement it in Excel. For the Python implementation, see our Historical Method VaR in Python tutorial.

What Is the Historical Method?

The historical method — also called historical simulation — is a non-parametric approach to calculating Value at Risk. Instead of assuming returns follow a normal distribution (like the parametric method) or generating random scenarios (like the Monte Carlo method), it uses the actual distribution of past portfolio returns to estimate potential losses.

Key Concept

The historical method lets the data speak for itself. It ranks all observed returns from worst to best and identifies the loss threshold at your chosen confidence level. No assumptions about the shape of the return distribution are needed — the historical data is the distribution.

This makes the historical method particularly appealing when returns exhibit fat tails, skewness, or other non-normal characteristics that the parametric method may fail to capture. For a high-level comparison of all three VaR methods, see our complete Value at Risk guide.

Video: VaR Historical Method Explained

How the Historical Method Works

The historical method follows a straightforward four-step process:

  1. Collect historical return data — Gather daily (or weekly/monthly) portfolio returns for a lookback window, typically 1 to 3 years. For daily VaR, one year equals approximately 252 trading days.
  2. Sort returns from worst to best — Arrange all observed returns in ascending order, so the largest losses appear first.
  3. Identify the percentile cutoff — Based on your confidence level, find the return at the appropriate percentile. For 95% confidence, this is the 5th percentile; for 99% confidence, it is the 1st percentile.
  4. Calculate dollar VaR — Multiply the absolute value of that percentile return by your portfolio value. The result is your VaR estimate.
Pro Tip

Use at least 1 year (252 trading days) of daily return data for a reliable VaR estimate. Shorter windows produce noisy percentile estimates, while 2 to 3 years of data captures a broader range of market conditions — including potential stress periods.

The Historical Method Formula

The historical method VaR can be expressed as:

Historical VaR Formula
VaRα = -Portfolio Value × Percentile(1-α)(R)
VaR at confidence level α equals the negative of the portfolio value times the (1-α) percentile of historical returns. Since the percentile return is negative (a loss), negating it yields a positive VaR amount.

Where:

  • α — confidence level (e.g., 0.95 for 95%)
  • R — the series of historical portfolio returns
  • Percentile(1-α)(R) — the (1-α) percentile of sorted returns (e.g., the 5th percentile for 95% confidence)

To find the exact observation, calculate the percentile position in the sorted data:

Percentile Position
Position = (1 – α) × n
Where n is the number of historical observations. If the position is not a whole number, interpolate between the two nearest sorted returns.

For example, with 252 daily observations and 95% confidence: Position = 0.05 × 252 = 12.6. This means the VaR falls between the 12th and 13th worst returns. In practice, you can either round up to the 13th worst return for a conservative estimate, or interpolate between the 12th and 13th values. Excel’s PERCENTILE.INC function handles this interpolation automatically, so the manual position calculation is most useful for building intuition about what the percentile represents.

Historical Method Example

$1M Portfolio — 95% 1-Day VaR

Given:

  • Portfolio value: $1,000,000
  • Lookback window: 252 daily returns (1 year of trading data)
  • Confidence level: 95%

Step 1: Sort all 252 daily returns from worst to best.

Step 2: Calculate the percentile position: 0.05 × 252 = 12.6 → look at the 13th worst return.

Step 3: The sorted tail of the return distribution might look like this:

Rank (Worst to Best) Daily Return
1st worst -4.8%
5th worst -3.2%
10th worst -2.5%
12th worst -2.2%
13th worst -2.1%
15th worst -1.9%

The 13th worst daily return is -2.1%.

Step 4: Calculate dollar VaR:

VaR = -$1,000,000 × (-0.021) = $21,000

Interpretation: Based on the past year of returns, there is a 95% probability that the portfolio will not lose more than $21,000 in a single trading day. Equivalently, on roughly 1 out of every 20 trading days, losses may exceed this threshold.

Historical Method vs Parametric Method

The two most commonly compared VaR approaches are the historical method and the parametric (variance-covariance) method. Each has distinct strengths depending on your portfolio and assumptions about return behavior.

Historical Method

  • No distributional assumptions required
  • Naturally captures fat tails and skewness
  • Simple and intuitive to implement
  • Limited by available historical data
  • Entirely backward-looking

Parametric Method

  • Assumes returns follow a normal distribution
  • Mathematically elegant closed-form solution
  • Fast computation with minimal data requirements
  • May underestimate tail risk
  • Scales easily to different time horizons

If your portfolio returns are approximately normal, the parametric method may be more efficient. However, if returns exhibit fat tails or skewness — as most equity portfolios do — the historical method provides a more realistic VaR estimate because it does not force data into a bell curve. In practice, many risk teams use both methods side by side: the parametric method for quick daily estimates and the historical method as a reality check that captures non-normal behavior.

How to Implement the Historical Method

Implementing the historical method in Excel requires only a few steps:

  1. Download historical prices — Use Yahoo Finance, Bloomberg, or another data provider to get daily closing prices for your portfolio’s assets over the lookback period.
  2. Calculate daily returns — For each day, compute the percentage return: (Pricet – Pricet-1) / Pricet-1. For a multi-asset portfolio, compute the weighted portfolio return for each day.
  3. Apply the PERCENTILE function — Use =PERCENTILE.INC(returns_range, 0.05) for 95% VaR, or =PERCENTILE.INC(returns_range, 0.01) for 99% VaR. This returns the percentile loss directly.
  4. Calculate dollar VaR — Multiply the absolute value of the percentile return by your portfolio value: =-portfolio_value * PERCENTILE.INC(returns_range, 0.05).
Pro Tip

Excel offers two percentile functions: PERCENTILE.INC (inclusive) and PERCENTILE.EXC (exclusive). For VaR calculations, PERCENTILE.INC is the more commonly used option — it includes the endpoints and matches the inclusive quantile convention used in many risk management applications.

For a full Python implementation using pandas and real market data, see our Historical Method VaR in Python tutorial. To learn all three VaR methods with video lessons and Excel workbooks, explore our Value at Risk course.

Video: Historical Method — Value at Risk (VaR) in Excel

Common Mistakes

Even experienced practitioners make these errors when applying the historical method:

1. Using too short a lookback window — With fewer than 100 observations, the 5th percentile is based on just 5 data points, producing an unreliable and unstable VaR estimate. Use at least 252 observations (1 year of daily data).

2. Ignoring regime changes and structural breaks — If your lookback window covers only a calm market period, the historical VaR will understate risk when volatility spikes. Conversely, including a crisis period that no longer reflects current conditions can overstate VaR.

3. Treating VaR as a worst-case scenario — VaR is a threshold, not a maximum loss. A 95% VaR of $21,000 means losses will exceed $21,000 about 5% of the time — and those exceedances can be much larger. Expected Shortfall (CVaR) captures the average magnitude of those tail losses.

4. Equal-weighting all observations — The standard historical method gives a return from 3 years ago the same weight as yesterday’s return. In rapidly changing markets, this can produce stale VaR estimates. Exponentially-weighted approaches address this by giving more weight to recent observations.

5. Mixing arithmetic and log returns — Using logarithmic returns for the percentile calculation but arithmetic returns for the dollar VaR conversion (or vice versa) produces inconsistent results. Pick one return convention and apply it consistently throughout.

Limitations of the Historical Method

Important Limitation

The historical method assumes the past is representative of the future. Unprecedented events — such as a pandemic, sovereign debt crisis, or flash crash not present in your data window — will not be captured by the model.

Backward-looking only — The historical method cannot anticipate risks that have never occurred in the sample. It provides no mechanism for forward-looking scenario analysis.

Sensitive to lookback period — Changing the lookback window (e.g., from 1 year to 3 years) can produce materially different VaR estimates. There is no universally “correct” window length.

May miss extreme tail events — If the historical sample does not contain a sufficiently severe drawdown, the VaR estimate will understate potential losses. This is especially problematic for short lookback periods.

Equal-weighting problem — All observations receive the same weight regardless of recency. A market crash from 2 years ago contributes equally to today’s VaR as yesterday’s return, even though current conditions may be very different.

For a forward-looking alternative that can model scenarios beyond historical experience, see the Monte Carlo VaR method.

Frequently Asked Questions

The industry standard is 1 to 3 years of daily return data (252 to 756 observations). Regulatory frameworks such as Basel guidelines have historically required a minimum of 1 year of data for internal VaR models. Longer lookback periods capture more extreme events and produce more stable estimates, but they may include outdated market regimes that no longer reflect current conditions. A common compromise is 2 years of daily data.

The historical method naturally captures fat tails because it uses actual observed returns rather than assuming a specific distribution. If extreme returns occurred during the lookback period, they will be reflected in the VaR estimate. This is a key advantage over the parametric method, which assumes normality and can underestimate the probability of large losses.

Yes. For a multi-asset portfolio, calculate the weighted portfolio return for each historical day using the current portfolio weights and each asset’s daily return. Then apply the percentile method to the resulting portfolio return series. This approach automatically captures correlations and co-movements between assets without requiring a separate correlation matrix.

Including a market crash in the lookback period will increase your VaR estimate, making it more conservative. This can be a benefit — it captures genuine tail risk that calm-period-only data would miss. However, it can also be a limitation if the crash occurred long ago and current market conditions are fundamentally different. Some practitioners use exponentially-weighted historical simulation, which gives more weight to recent observations while still capturing older extreme events.

VaR tells you the loss threshold at a given confidence level — the level of loss that will not be exceeded a certain percentage of the time. Expected Shortfall (CVaR) goes further by measuring the average loss in the worst cases beyond the VaR threshold. For example, if 95% VaR is $21,000, Expected Shortfall answers: “When losses do exceed $21,000, how bad are they on average?” This makes ES a more complete measure of tail risk.

The historical method is the simplest and most transparent of the three VaR calculation approaches. Its strength lies in making no assumptions about how returns are distributed — it uses real data as-is. To compare it with alternatives, see our guides on the parametric method and Monte Carlo method, or take the full Value at Risk course to master all three approaches with video lessons and Excel workbooks.

Disclaimer

This article is for educational and informational purposes only and does not constitute investment advice. The example calculations use hypothetical data for illustration. Actual VaR estimates depend on your specific portfolio, data source, lookback period, and methodology. Always conduct your own analysis and consult a qualified risk management professional before making investment decisions.